Expect the Expected: Approximating the Caliber of Possession Using Shot Quality James McCorriston Connor Reed
[email protected] [email protected] Abstract The NHL has experienced rapid growth in analytical metrics and advanced statistics in recent years. While popular statistics like Fenwick and Corsi act as good approximations for puck possession, they are limited in what they tell about scoring opportunities as they do not consider shot quality. In this study, we consider shot distance as an approximation of shot quality, and we combine Fenwick and NHL play-by-play shot distance data to develop a series of new statistics: Expected Goals (xGoals), Expected Differential (xDiff), and Goals-Above-Expected (GAE) for skaters, as well as Expected Save Percentage (xSv%) and Adjusted Save Percentage for goaltenders. As a basis for these new metrics, we first show that shot distance serves as a good approximation for shot quality, and that we can reverse-engineer scoring probabilities for each shot taken by a player. The concept of approximating shot quality is extended to analyze the performance of players, teams, and goaltenders. Using NHL play-by-play data from the 2007-08 season to the 2014-15 season, we show that xGoals are the best indicator of how many goals a player should be scoring, and we show that it stays more consistent for an individual from year-to-year than other comparable statistics. Finally, we show that on a single-game resolution, xGoals are the best indicator for which team should have won a particular game. The novel set of metrics introduced in this paper offer a more reliable and indicative tool for assessing the ability of skaters, goaltenders, and teams and provides a new basis for analyzing the game of professional hockey.